I put too much time into this link to the project (Vitanova) - github.com/4t-... Bibites project - / @thebibitesdigitallife LifeEngine project - / @emergentgarden Discord - / discord
I think using simplified swimming hydrodynamics like what Darwin Pond and Swimbots used would be a better way to have faster swimming that isn't just "more throttle". It also enables more interesting combat movements. Even 3 segments already work wonders Love the polished presentation
That is something I am planning to add but I just wanted to first make a working prototype before I started to add more features. (Though expect something like that to be added hopefully “soon”)
@@revimfadli4666 Gene Pool is much more ambitious and more refined. Darwin Pond is a very old piece of software, and so is difficult to get running well, and its simulation was also much simpler. This simplicity though meant that it was within your reach to tweak the sliders on your own to create and tune swimmers. In Gene Pool, it is much more difficult to create a custom swimbot, but the possibilities and variety of forms is absolutely huge!
oh it’s really cool that you do neural and physiological evolution. the only other project i’ve seen do that is the bibites and a mod for the life engine
the millisecond i clicked on this video i was like "isn't this basically just the bibites by someone else?" glad to see you called that out, or got inspired by it, or both
7:03 - You can just use atomic operations and that way you can safely add. Also applications like this should be relatively easily parallelizable. Just throwing parallelism out of the window is not wise, I mean the only reason why we can train or even run large neural network is because we can parallelize most of the operations we do on them.
I always daydreamed about something like this and I finally found it, I haven't watched the whole video so i don't know if you are also going to allow the creatures to change their shapes and also evolve visually. But that will probably be too hard :(
Yeah having them evolve their bodies beyond basic sliders (size, speed, colour, etc) and have them actually have limbs that change will probably be one of the hardest features to add because there are just way too many factors to consider (how the bodies will be structured, how they evolve, how the network will control it if at all, just to name a few) but hopefully it will become a feature one day as it would lead to many more different possible forms of artificial life.
For mutations you should only flip one bit. Yes you'll get smaller changes but you'll get fewer changes that just die. Oh you improved the mutation stuff a lot already (-:
I really want to add this to a 3d God simulator if I ever get good at programming. Alas it will probably take 5 years to learn enough programming to make it (especially if I do not binge programming as much as I should)
I did actually consider going 3D (and technically with a big code rewrite it still could) but 2D is just a lot easier to deal with in both writing code and also performance. But a 3D version might be amazing in VR and would probably just cause me to just do nothing and watch life go by and evolve with the occasional meddling in what happens. Programming is not as hard as you might think to learn as most of it isn't coding but problem solving, you can learn how to use a programming language in a pretty short time and once you learn one language the next becomes even easier as most of them follow the same rough pattern. You get good at programming like any other skill by practicing and working on projects (start small obviously but move up in complexity). There are plenty of tutorials online on different languages (probably best to start with something popular like Python, Java, JS, or even C/C++) to help you get started, I might even try and make one of my own one day.
You should try to add friction to the system, make it so that turning to any cardinal position spends the same amount of weight and then mutate this parameter. Also, when you talk about evolution, you say that the faster you are, more energy you spend, but that is not what actually happens in nature. Take a look at the diference in metabolism between an elephant and a mouse. Moving more mass takes exponentially more energy as the scale goes up, unless you are in a fluid. You can notice that when they are small they moce very slowly and faster as they get bigger
I was looking over the source and I'm sure I seen some sort of resistance in there. Although I'm a fairly shit python programmer, so I was just sort of scratching my head looking at this C++ project. lol
If you go on other learning methods it won't be simulating evolution anymore... Also we couldn't see anything in the time lapse... But keep it up. I love the Bibites project and will come by to see your version of it.
The creature's neural network has 12 inputs, A constant on, their X pos, their Y pos, their rotation, their speed, their energy, their health, their age, closest food distance and rotation, and closest creature distance and rotation. All of these inputs range from 1 to -1 They can aim pretty easily as to do that they only need to connect the food rotation node to the output node that controls rotation. That only needs one lucky mutation which with lots of creatures is bound to happen at least once. So when the food rotation node says that there is food to the right, the rotation node it is connected to will activate and it will turn right, very simple but it works very well.
I didnt talk about it in the video but I saw that some of the creatures evolved to use their position to turn away from the edge of the world when they got too far out to stop them from moving into the abyss never to be seen again and eventually die from starvation. The first time I saw that I was honestly pretty surprised as that was the first smart thing I saw the creatures do beyond turning towards food.
This project is pretty cool.
I guess you could say, it's pretty NEAT!
wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool wow that's pretty cool
I think using simplified swimming hydrodynamics like what Darwin Pond and Swimbots used would be a better way to have faster swimming that isn't just "more throttle". It also enables more interesting combat movements. Even 3 segments already work wonders
Love the polished presentation
That is something I am planning to add but I just wanted to first make a working prototype before I started to add more features.
(Though expect something like that to be added hopefully “soon”)
@@scasz going with a minimum viable prototype first is a solid principle indeed. Looking forward to the next vid
Darwin Pond was an awesome little program! Hard to run well on modern systems, though.
@@HansLemursonwhat do you think about its sequel, Gene Pool?
@@revimfadli4666 Gene Pool is much more ambitious and more refined. Darwin Pond is a very old piece of software, and so is difficult to get running well, and its simulation was also much simpler. This simplicity though meant that it was within your reach to tweak the sliders on your own to create and tune swimmers.
In Gene Pool, it is much more difficult to create a custom swimbot, but the possibilities and variety of forms is absolutely huge!
oh it’s really cool that you do neural and physiological evolution. the only other project i’ve seen do that is the bibites and a mod for the life engine
the millisecond i clicked on this video i was like "isn't this basically just the bibites by someone else?"
glad to see you called that out, or got inspired by it, or both
7:03 - You can just use atomic operations and that way you can safely add. Also applications like this should be relatively easily parallelizable. Just throwing parallelism out of the window is not wise, I mean the only reason why we can train or even run large neural network is because we can parallelize most of the operations we do on them.
I always daydreamed about something like this and I finally found it, I haven't watched the whole video so i don't know if you are also going to allow the creatures to change their shapes and also evolve visually. But that will probably be too hard :(
Yeah having them evolve their bodies beyond basic sliders (size, speed, colour, etc) and have them actually have limbs that change will probably be one of the hardest features to add because there are just way too many factors to consider (how the bodies will be structured, how they evolve, how the network will control it if at all, just to name a few) but hopefully it will become a feature one day as it would lead to many more different possible forms of artificial life.
Check out the bibites!
This video is pretty neat
Nice. I like this.
This is awesome man. I really enjoyed that.
Welp this guy is going to go viral one day time to become an OG to a up and coming legend.
the style of the video and the pacing of information indeed are really nice
nifty way you do graphics and diagrams. simple but i like it :)
For mutations you should only flip one bit. Yes you'll get smaller changes but you'll get fewer changes that just die. Oh you improved the mutation stuff a lot already (-:
This is so freaking cool
the tiny dots on a black backgroudn arent the easiest to see :c
alas, i dont really have any ideas to fix that
You should implement some tools for the creatures to communicate with each other
I really want to add this to a 3d God simulator if I ever get good at programming.
Alas it will probably take 5 years to learn enough programming to make it (especially if I do not binge programming as much as I should)
I did actually consider going 3D (and technically with a big code rewrite it still could) but 2D is just a lot easier to deal with in both writing code and also performance. But a 3D version might be amazing in VR and would probably just cause me to just do nothing and watch life go by and evolve with the occasional meddling in what happens.
Programming is not as hard as you might think to learn as most of it isn't coding but problem solving, you can learn how to use a programming language in a pretty short time and once you learn one language the next becomes even easier as most of them follow the same rough pattern. You get good at programming like any other skill by practicing and working on projects (start small obviously but move up in complexity). There are plenty of tutorials online on different languages (probably best to start with something popular like Python, Java, JS, or even C/C++) to help you get started, I might even try and make one of my own one day.
HOW IS THERE ONLY 40 COMMENTS BEST VIDEO ON UA-cam
This video is sick
I thought this video would of bad 100k views or something dammn your underrated
Thank you, realized where I was going wrong with neurals.
Interesting
You should try to add friction to the system, make it so that turning to any cardinal position spends the same amount of weight and then mutate this parameter.
Also, when you talk about evolution, you say that the faster you are, more energy you spend, but that is not what actually happens in nature. Take a look at the diference in metabolism between an elephant and a mouse. Moving more mass takes exponentially more energy as the scale goes up, unless you are in a fluid.
You can notice that when they are small they moce very slowly and faster as they get bigger
I was looking over the source and I'm sure I seen some sort of resistance in there. Although I'm a fairly shit python programmer, so I was just sort of scratching my head looking at this C++ project. lol
Awesome
If you go on other learning methods it won't be simulating evolution anymore...
Also we couldn't see anything in the time lapse... But keep it up. I love the Bibites project and will come by to see your version of it.
Awesome project ! What are the input exactly? How is it so easy for them to aim to food?
The creature's neural network has 12 inputs, A constant on, their X pos, their Y pos, their rotation, their speed, their energy, their health, their age, closest food distance and rotation, and closest creature distance and rotation. All of these inputs range from 1 to -1
They can aim pretty easily as to do that they only need to connect the food rotation node to the output node that controls rotation. That only needs one lucky mutation which with lots of creatures is bound to happen at least once. So when the food rotation node says that there is food to the right, the rotation node it is connected to will activate and it will turn right, very simple but it works very well.
@@scasz ok thanks for the explanation! Did you see any creature using X and Y in an interesting way ?
I didnt talk about it in the video but I saw that some of the creatures evolved to use their position to turn away from the edge of the world when they got too far out to stop them from moving into the abyss never to be seen again and eventually die from starvation. The first time I saw that I was honestly pretty surprised as that was the first smart thing I saw the creatures do beyond turning towards food.
Can you pls make a tutorial on how to do it?
:)
Real
You wanted 60 frames in 1 second? No, the game decided it should be 1 frame per 60 seconds.
horse
no